FExM simplifies basic fuzzing pipelines. As result of years of practical fuzz testing, it provides a best effort approach able to get running quickly and finding bugs in most applications. Given its fully automated nature, it can be used to fuzz complete software repositories and distributions as well as single programs.
These instructions guide you through your first fuzz test using FExM.
Right now, FuzzExMachina runs natively on Ubuntu Linux (18.04).
If you just want to try it out, we provide a vagrant vm with the correct environment for easy setup. The Dockerfile provided might also work but at this point is pretty untested.
To use Vagrant please install Vagrant, the Virtualbox provider and nfs.
- Software
- vagrant (Tested with v2.1.2)
- virtualbox (Tested with v5.2.14)
- Recommended Vargant Plugins (install with
vagrant plugin install <name>
)- vagrant-disksize -> More space inside Vagrant disks
- vagrant-winnfsd -> Shared folders via NFS (Windows only)
- [network file system] sudo apt-get install nfs-common
- Hardware
- 8GB RAM+
- Sufficient disk space (== a lot)
- 8 logical CPUs (or adapt the Vagrantfile manually)
To get running using Vagrant, simply
git clone git@github.com/fgsect/fexm.git
cd fexm && vagrant up
vagrant ssh
For a manual setup (intrusive!), run ./fexm/scripts/setup.sh
. It performs the following steps:
- Install docker
- Install RabbitMQ
- Creates the FExM virtualenv in
~/.virtualenvs
(for virtualenvwrapper) and installs the packages from./fexm/requirements.txt
inside venv
First, you will need to acquire seeds that will then be fed into the FExM.
For this, FExM provides a github crawler.
To do that, create an auth token at GitHub.
Afterwards, run
./fexm/fexm crawl -a <GitHub AUTHTOKEN> -o <PathToSeeds>`
Inside vagrant, you can simply run:
fexm crawl -a <GitHub AUTHTOKEN>
After creating the seeds, it's time to fuzz. FExM is configured via json. The following is a commented version of a json configuration file:
{
"name": "top500", // The name of our fuzzing run
"fuzz_manager": "pacman", // What fuzz manager to use (inside ./)
"packages_file": "/fexm/examples/top500.txt", // for pacman fuzzer: top500.txt
"base_image": "pacmanfuzzer" // the base docker image to use
}
- Set up pacmanfuzzer docker image
Pacmanfuzzer is an example fuzzer that will fuzz the arch upstream repositories.
fexm init pacmanfuzzer
Verify that the docker image has been set up in the previous step
$ docker images -a | grep pacmanfuzzer
pacmanfuzzer latest 489673de5625 About a minute ago 2.98GB
We provide an example json that will fuzz the top500 tools from the Arch repos as top500.json
.
fexm fuzz ./examples/top500.json
This will then spawn the 2. Another one for the example python script. Your results will be stored in the out directory.
To display the results in the dashboard, open http://localhost:5307 in your browser.
If you liked the demo, check out the source code of FExM. Forking and starring is highly welcome! We would like FExM to be free and open source, and most importantly useful to you. Since we have limited resources to maintain code and add new features, we encourage you to send pull requests for bug fixes and new features :-)
- To connect to the docker container do:
docker run -ti --entrypoint /bin/bash --privileged pacmanfuzzer
--privileged is need for strace and asan.
Now you are inside the docker!
To build a package with afl, do:
aflize package
or even better, use the python script:
fexm/configfinder/builder_wrapper.py -p <package>
To fuzz, just use afl-fuzz as you would normaly do. Seeds are in the /tmp/seeds/
, dictionaries are in /fuzz_dictionaries/dictionaries/
To mount your current path as volume, do:
docker run -v "$(pwd):/results" -ti --entrypoint /bin/bash --privileged pacmanfuzzer
To use the detailed tooling, spawn up a celery instance:
cd /fexm/celery_tasks
celery -A tasks
You can now use the tools (for more information call -h
on each tool):
python fexm/tools/inference_manager_pacman.py -cd syncdir -di pacmanfuzzer
pyhton fexm/fuzz_manager/fuzz_manager_round_robin.py -cd syncdir -di pacmanfuzzer -t <afl -t option>
# Inferring input vectors
To infer the input vector for a given binary, execute:
$ /fexm/configfinder/config_finder_for_binary.py -b tcpdump -s /seeds
[...]
#########################################################################
Input vector for tcpdump is most likely:
-nr @@
/seeds/pcap_samples
The tool will automatically figure out when QEMU is required.
Q: I get pull access denied for pacmanfuzzer, repository does not exist or may require 'docker login'
when trying to fuzz.
A: You need to run fexm init pacmanfuzzer
once before fuzzing.
See also the list of contributors who participated in this project.
This project is licensed under the APACHE v2 License - see the LICENSE.md file for details.
- rc0r for AFL-Utils
- jfoote for exploitable
- Lots of groundwork on preeny was done by @zardus
- @lcamtuf for AFL (obviously)
- Some initial work for the pcap parser was stolen (with permission) from Ben Stock (@kcotsneb)
- The GitHub crawler makes use of sourcecode by Tommi Unruh
- https://github.com/d33tah/ for the great aflize script.
- So many more...